🧠 DeepCaps_Pytorch

PyTorch implementation and training of DeepCaps: Going Deeper with Capsule Networks

🚀 Overview

DeepCaps_Pytorch is a PyTorch implementation of DeepCaps: Going Deeper with Capsule Networks, based on the paper "DeepCaps: Going Deeper with Capsule Networks" (Rajasegaran et al., 2019). See: https://arxiv.org/abs/1904.09546

This project adapts the DeepCaps architecture for applications such as galaxy morphology classification (using SDSS / Galaxy Zoo data) as part of an academic research investigation.
It demonstrates how Capsule Networks can be extended to deeper architectures and compared against CNN-based baselines.

✨ Features

• 🧩 Modular implementation of DeepCaps architecture in PyTorch
• 📈 Training, evaluation, and prediction scripts
• 🪐 Dataset support for SDSS / Galaxy Zoo images
• ⚙️ Configuration system (cfg.py) for hyperparameter tuning
• 📊 Visualization tools for accuracy and loss curves
• 🧠 Easily extendable to new datasets and domains

🧠 Tech Stack

• Language: Python 3.10+
• Framework: PyTorch
• Visualization: Matplotlib, Seaborn
• Utilities: NumPy, tqdm, Pillow

📁 Project Structure

├── acc_plot.py           # Plot accuracy/loss curves
├── cfg.py                # Configuration / hyperparameters
├── helpers.py            # Utility functions
├── load_data.py          # Dataset loader base
├── load_data_sdss.py     # SDSS-specific data loader
├── model.py              # DeepCaps architecture
├── plot.py               # Visualization utilities
├── predictor.py          # Inference and prediction
├── train.py              # Model training script
├── requirements.txt
└── README.md

⚙️ Installation

Clone the repository and install dependencies:

git clone https://github.com/Commit2Cosmos/DeepCaps_Pytorch.git
cd DeepCaps_Pytorch
pip install -r requirements.txt

If using a GPU, ensure PyTorch is installed with CUDA support. You can verify this by running:

python -c "import torch; print(torch.cuda.is_available())"

💻 Usage

🏋️ Training

To train the DeepCaps model, simply run:

python train.py

You can modify hyperparameters (epochs, learning rate, dataset paths, etc.) inside cfg.py.

🔮 Prediction / Inference

Once trained, you can run inference using:

python predictor.py

📊 Examples / Results

🧩 How to Use with Your Own Dataset

1. Update cfg.py — modify dataset paths, batch size, and training hyperparameters.
2. Add a new loader — write a new dataset class in load_data.py following the torch.utils.data.Dataset interface.
3. Edit train.py — point the dataset reference to your new loader.
4. Run training — DeepCaps will adapt to the new data dimensions and classes.
5. Evaluate and visualize — use acc_plot.py and plot.py to generate metrics.

📚 References

1. Rajasegaran, J., Jayasundara, V., Jayasekara, S., et al. (2019). DeepCaps: Going Deeper with Capsule Networks. arXiv:1904.09546
2. Original TensorFlow Implementation
3. PyTorch Reference Implementation
4. Galaxy Zoo dataset